I think the insane difficulty of going from zero to one is what attracts people to the MVP idea:. 1. launch a series of half-baked ideas.2. see what sticks.3. build what stuck. But can only play this “throw shit on the wall” strategy if you’re flexible on the idea. If you want.

What will “AI taking jobs” look like?.

What do you want to know?.

Follow-up shower thought. When people imagine AI “takes jobs”, what are they imagining?. (A) replace domain specialists with low-skilled labor that writes stream-of-consciousness, poorly-formatted, vaguely-instructed zero-shot prompts to a consumer chatbot like ChatGPT?. (B).

I wonder if using professional / academic jargon contributes to o3 and GPT-5 seeming “smarter” too. Whereas 4o would tend to phrase things in a more pedestrian manner.

ChatGPT 5 Thinking seems to like using heavy jargon (without explanation or handholding, assuming the user is an expert). o3 was fond of it, but GPT-5 seems to lean into it even more, to the point of using abbreviations. Like “dd” for “drawdown” when referring to investments.

They’re trying to balance between “regular people” and people who have done the legwork to understand the differences between models. But supposing that’s already covered by “Auto”, I think the rest of the options should just reveal what model is being used under the hood. Like

In trying to make it simpler, I fear OpenAI has made it more confusing to understand what model you’re using. In ChatGPT there are now different models you can select than the ones available in the API, and it’s not clear which ones map to which. Is “ChatGPT-5 Thinking” GPT-5.

But: I’ve read of reports from those who’ve experienced a step change improvement from GPT-5 to Sonnet 4 when they changed their prompting style. So my next move is to read and internalize the OpenAI cookbook for GPT-5 prompting.

Managed to get it done in a few passes but it wasn’t exactly a walk in the park. So far, at least for this kind of task, GPT-5 feels kind of like working with codex-1 or Sonnet 4. It doesn’t feel like a step change in capability here, yet.

Not going too great. Gave me an invalid postcss.config.js file. Now trying to ask it to just follow the shadcn/ui docs and init again. For context: this project is just a small single-page prototype, with no cursorrules, so it’s interesting to see what Cursor/GPT-5 does out

Siccing GPT-5 in Cursor on a pretty “easy” task: upgrade Tailwind v3 to v4. In my experience, these kinds of tasks tend to be easy for humans but difficult for LLMs (even with carefully curated context and instructions) . Let’s see how it handles it, since GPT-5 is supposedly

If the interface is different it’ll usually attract a different crowd that’s always been technically-inclined but simply struggling with the abstraction presented by the interface. When someone figures out how to unlock a new kind of interface for a task previously accomplished.

The point is that syntax and interface isn’t what stops “non-technical people” from casually using an interface for interacting with a technology. Being “technical” is about understanding the full range of knobs & dials that you can turn to achieve a particular outcome. Turns out.

This feels like a replay of SQL and Markdown. SQL: invented as a language to make querying databases “easy” for non-technical business users. Markdown: invented as a markup language to make blogging and commenting on blog posts “easy” for non-technical authors and readers. SQL.

Things to sidestep now, thanks to AI:. em dashes, “certainly”, “delve” “it’s not X, it’s Y” in writing. Any sort of purple or cool-tinted gradients in design. Lucide icons, system fonts, built-in Tailwind CSS colors. “Fade-in and translate up from -1px” CSS transitions that.

Finally OpenAI's model naming makes sense now. gpt-5-pro.gpt-5-mini.gpt-5-nano.gpt-oss-120b.gpt-oss-20b.gpt-image-1. Hopefully they don't go and spoil it 2 months later by launching gpt-5.1o-turbo or something weird like that.

There was a brief period ~10 years ago when tech companies cared about UX and I miss it so much. Now everything is flashier, but it sucks. The jobs “UX designer” and “interaction designer” are more or less gone. While it’s become easier to ship good UX, it’s also become quite.

Replace “game design” with any creative endeavor and it kind of still fits. Analysis of prior work doesn’t lead to understanding of how to make good stuff. Analysis only helps develop taste & appreciation. Only in trial and error do you learn how to create something well designed.